The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requests full shell access to automate project scaffolding and validation tasks.
Evidence: SKILL.md grants Bash(cmd:*) permissions.
Evidence: references/plugin-creation-process.md describes executing npm run sync-marketplace and a local validation script ./scripts/validate-all.sh during the plugin generation flow.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it incorporates untrusted user input into generated files which could influence downstream agents or tools.
Ingestion points: User-provided plugin name, description, and keywords captured during the 'generate', 'create', or 'scaffold' triggers defined in SKILL.md.
Boundary markers: Absent. There are no instructions or delimiters designed to isolate user input from the generated file structure.
Capability inventory: The skill possesses Write, Grep, and Bash(cmd:*) tools (as seen in SKILL.md), which could be misused if instructions hidden in generated files are subsequently executed.
Sanitization: While the skill mentions using jq for JSON syntax validation and a custom validation script, there is no evidence of content sanitization for natural language fields like descriptions (described in references/plugin-creation-process.md).

plugin-creator